Recommendation Algorithm

The types of recommendation algorithms can be broadly classified into three categories

With the advent of the Big Data era, various forms of data reflecting consumers' behaviors and interests are being generated. Through this data, an environment is being created to predict and support tourists' decision-making processes. Furthermore, the increased accessibility of information for tourists has provided convenience in decision-making, but the overwhelming amount of information makes it challenging to process all of it. To alleviate this burden, businesses are incorporating "recommender systems" that assist tourists in their decision-making process.

Particularly, online recommender systems have become widely utilized in recent industries. These systems predict preferences based on users' past interests, purchases, and other activities, enabling them to suggest optimal products or services. As a result, companies can provide personalized recommendations to users, enhancing their overall experience and satisfaction.

Figure 1. The Operational Concept of How the Recommendation Algorithm Works

Determinants of Predictive Power of Tourist Recommendation Algorithms(2022)

There are three types of recommendation algorithms

Content-based filtering: This algorithm utilizes only the content to provide recommendations. It analyzes the content that a user likes and suggests similar content based on that information. For example, if a user enjoys listening to the song "All I want for Christmas is you," the algorithm may recommend other songs related to winter and Christmas.

Collaborative filtering: Collaborative filtering collects preference information from other users to predict a user's interests. The CF model uses interactions between users and items, assuming that users with similar preferences will have similar liking towards certain items. For instance, if people who like carols commonly show high preference for fantasy movies, the algorithm may recommend "Harry Potter" to a user with a liking for carols.

Hybrid-based filtering: Hybrid-based filtering combines both content-based and collaborative filtering approaches to provide more accurate recommendations. It takes advantage of the strengths of both methods to enhance the recommendation process and offer personalized suggestions to users..

Figure 2. Conceptual Framework for Understanding recommend system

wikidocs.net/recommendation system

Collaborative filtering can be divided into memory-based and model-based approaches. The memory-based approach is the most traditional method and operates based on similarity. When using user similarity as a criterion, it is called user-based recommendation, and when using item similarity, it is referred to as item-based recommendation. While memory-based approaches rely on existing data for recommendations, model-based approaches calculate hidden feature values of users or items through machine learning. Model-based approaches encompass various methods, such as Latent Factor using latent factors, Classification/Regression-based methods, and more recent fusion models.

Fusion methods include Factorization Machine, which combines characteristics of Latent Factor models and Classification/Regression models, as well as Neural Collaborative Filtering, an extension of Latent Factor models using deep learning.

Tourist attraction recommendation algorithms

Tourist attraction recommendation algorithms can be categorized into survey-based, location-based, and content-based recommendation algorithms. The survey-based recommendation algorithm derives a personalized list of recommended tourist attractions based on survey information targeting. Despite the advantage of obtaining various information through diverse questions, survey-based approaches suffer from the limitation that the collected data are the respondents' "perceived" information, thus lacking the value of objective measurements.

Location-based recommendation algorithms, on the other hand, utilize latitude and longitude information to recommend tourist attractions near the current location of the tourist. This type of service is commonly found in websites and applications, such as the "Nearby Attractions" feature in the web services provided by the Korea Tourism Organization or GPS navigation systems. Additionally, real-time weather information is also employed in location-based tourist attraction recommendation algorithms

Figure 3. The Calculation Formulas for User and Item Vectors are As Follows

Collaborative filtering recommendation methods, which can be categorized into user-based and item-based approaches, utilize customer preferences or liking information. These methods construct groups of customers with similar preferences as the basis for recommending tourist attractions. By comparing different groups, tourist attractions from groups with higher similarity are recommended to new customers. Specifically, for customers who have not visited a particular attraction, the algorithm measures their preferences and similarity based on the preferences of customers who have visited that attraction. It then recommends attractions that have been favored by customers with similar preferences.

While content-based recommendation focuses on recommending attractions based on their attributes, collaborative filtering, as its name suggests, involves comparing customer preferences. The drawback of collaborative filtering lies in the need for sufficient tourist data to compare preferences . However, this study overcomes this limitation by utilizing big data from tourists.

In this study, for the sake of simplicity in interpreting the results, precision and recall values derived from the deep learning recommendation algorithm were visualized as a 2D curve in the first quadrant. The precision-recall curve for preference-based tourist attraction recommendation starts at the top-left value of 1 and gradually decreases to the bottom-right value of 1. Additionally, it can be observed that the curve remains close to the diagonal line connecting the points of value 1. This indicates that the precision-recall curve closely follows the line representing the value of 1, demonstrating a high predictive power for this recommendation algorithm.

Figure 4. Preference-based Tourist Attraction Recommendation Algorithm Precision-Recall.

The "sequence-based recommendation algorithm" proposed in this research is expected to contribute significantly to the technological advancement by overcoming the limitations of existing tourist attraction recommendation techniques, such as distance-based and content-based algorithms, and reflecting tourists' preferences to enhance satisfaction. Moreover, it is anticipated to provide a perspective from the marketing dimension of tourism, moving away from the sole focus on accuracy in the technological development of recommendation algorithms and contributing to the interdisciplinary development between natural sciences and social sciences.

Most importantly, the sequence-based approach presented in this research offers an alternative perspective that goes beyond the dichotomy of distance and content, reflecting tourists' preferences according to their order. This alternative perspective allows for a theoretical examination of recommendation algorithms that consider changes in tourists' consumption behavior, through the integration of technological-focused recommendation algorithm research and the management perspective of marketing.

In related interdisciplinary research, there are examples of applying text mining technology to tourism studies. These previous studies also emphasize the value of developing recommendation algorithms that utilize user opinions. In the same context, the collaborative filtering recommendation algorithm proposed in this study simultaneously utilizes tourists' preference information and evaluations from other tourists (ratings and reviews) for individual attractions. By incorporating varying weights to different criteria, it is expected to contribute to improving the alignment between recommended attractions and customer preferences.

References

Yang, O. S. and Shin, S. R. (2022) ‘Determinants of Predictive Power of Tourist Recommendation Algorithms : Focusing on the Hierarchical Ranking Theory between Preference and Distance,'The Korean Society of Management Consulting', 22(6), pp. 329-340.

Aaker, D. A. (1996) Building Strong Brand, New York, NY: The Free Press.

Lee, H., Chung, N. and Nam, Y. (2019) 'Do online information sources really make tourists visit more diverse places? based on the social networking analysis', Information Processing

& Management, 56(4), pp. 1376-1390.

Hwang, Y. H. and Fesemaier, D. R. (2011) ‘Unplanned tourist attraction visits by travellers’,

Tourism Geographies, 13(3), pp. 398-416.

Lemmon, M. L. and Zender, J. F. (2010) ‘Debt capacity and tests of capital structure theories’, Journal of Financial and Quantitative Analysis, 45(5), pp. 1161-1187.